AAAI 2024 Papers — Page 6
AAAI Conference on Artificial Intelligence · 2331 papers
Delegation-Relegation for Boolean Matrix Factorization
Florent Avellaneda (University of Quebec at Montreal), Roger Villemaire (University of Quebec at Montreal)
OptimizationTabular
🎯 What it does: This paper proposes two operators, delegation and relegation, to simplify Boolean matrices, thereby reducing the number of 1 elements that need to be factored while maintaining or guaranteeing the minimum rank, significantly lowering the solving time for constraint-based BMF.
Deletion-Robust Submodular Maximization with Knapsack Constraints
Shuang Cui (University of Science and Technology of China), He Huang (Soochow University)
Recommendation SystemOptimizationGraph
🎯 What it does: A streaming algorithm is proposed for the robust submodular function maximization (RSK) problem, supporting both non-monotonic and monotonic submodular functions, achieving approximately optimal subset selection under budget constraints.
Delivering Inflated Explanations
Yacine Izza (National University of Singapore), Joao Marques-Silva (IRIT CNRS)
Explainability and InterpretabilityComputational EfficiencyTabular
🎯 What it does: This paper proposes and implements 'inflated explanations' by expanding the range (or set) of feature values in traditional suspicious explanations to provide more informative explanations.
Delving into Multimodal Prompting for Fine-Grained Visual Classification
Xin Jiang (Nanjing University of Science and Technology), Zechao Li (Tongji University)
ClassificationTransformerPrompt EngineeringContrastive LearningImageMultimodality
🎯 What it does: A Fine-Grained Visual Classification method based on multimodal prompts, MP-FGVC, is proposed. It utilizes the cross-modal descriptive ability of the CLIP model to capture fine-grained differences in subcategories and achieve cross-modal collaborative reasoning through visual prompts (SSVP), text prompts (DATP), and a visual-language fusion module (VLFM), employing a two-stage training strategy for efficient fine-tuning.
DenoSent: A Denoising Objective for Self-Supervised Sentence Representation Learning
Xinghao Wang (Fudan University), Xipeng Qiu (Fudan University)
Representation LearningTransformerContrastive LearningText
🎯 What it does: A self-supervised representation learning framework based on sentence denoising, called DenoSent, is proposed;
Dense Projection for Anomaly Detection
Dazhi Fu (Chinese University of Hong Kong), Jicong Fan (Hefei University of Technology)
Anomaly DetectionImageTabular
🎯 What it does: A local density maximization method based on neural networks (DPAD) is proposed, which performs unsupervised anomaly detection by learning local high-density representations and utilizing kNN density scoring.
Density Matters: Improved Core-Set for Active Domain Adaptive Segmentation
Shizhan Liu (Shanghai Jiao Tong University), Weiyao Lin (Shanghai Jiao Tong University)
SegmentationDomain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A density-aware core set selection method is proposed for active domain adaptive semantic segmentation.
Dependency Structure-Enhanced Graph Attention Networks for Event Detection
Qizhi Wan (Jiangxi University of Finance and Economics), Dexi Liu (Jiangxi University of Finance and Economics)
Graph Neural NetworkSupervised Fine-TuningText
🎯 What it does: Proposes an event detection framework enhanced by dependency structures, converting dependency relationship edges into nodes to construct a Dependency Relationship Graph (DRG), which is jointly encoded with the traditional Token Dependency Graph (TDG) to enhance the hierarchy of core argument nodes;
DePRL: Achieving Linear Convergence Speedup in Personalized Decentralized Learning with Shared Representations
Guojun Xiong (Stony Brook University), Jian Li (Stony Brook University)
OptimizationFederated LearningRepresentation LearningConvolutional Neural NetworkImage
🎯 What it does: The DePRL algorithm is proposed to achieve decentralized personalized learning through the sharing of low-dimensional representations and collaborative training of individual local heads.
Depth-Guided Robust and Fast Point Cloud Fusion NeRF for Sparse Input Views
Shuai Guo (Shanghai Jiao Tong University), Li Song (Shanghai Jiao Tong University)
GenerationData SynthesisCompressionComputational EfficiencyNeural Radiance FieldPoint Cloud
🎯 What it does: A deep-guided fast point cloud fusion NeRF is designed for viewpoint synthesis with sparse view inputs.
DeRDaVa: Deletion-Robust Data Valuation for Machine Learning
Xiao Tian (National University of Singapore), Bryan Kian Hsiang Low (National University of Singapore)
Data-Centric LearningTabular
🎯 What it does: This paper proposes a deletion-robust data valuation method, DeRDaVa, aimed at data deletion scenarios, and extends it to Risk-DeRDaVa to accommodate model owners who are risk-averse or risk-seeking.
DeS3: Adaptive Attention-Driven Self and Soft Shadow Removal Using ViT Similarity
Yeying Jin (National University of Singapore), Robby T. Tan (National University of Singapore)
RestorationTransformerDiffusion modelImage
🎯 What it does: Proposes DeS3, a single image shadow (hard, soft, self-shadow) removal method based on diffusion models, adaptive attention, and ViT similarity loss.
Descanning: From Scanned to the Original Images with a Color Correction Diffusion Model
Junghun Cha (Kyung Hee University), Sung-Ho Bae (Adobe Research)
Image TranslationRestorationDiffusion modelImage
🎯 What it does: This paper proposes a 'Descanning' task that recovers original digital images from scanned images, and based on this, introduces a new deep learning framework called DescanDiffusion.
Designing Biological Sequences without Prior Knowledge Using Evolutionary Reinforcement Learning
Xi Zeng (Northwestern Polytechnical University), Jiajie Peng (Northwestern Polytechnical University)
OptimizationConvolutional Neural NetworkRecurrent Neural NetworkReinforcement LearningBiomedical Data
🎯 What it does: A novel evolutionary reinforcement learning framework named ERLBioSeq is proposed for designing biological sequences such as DNA, RNA, and proteins under conditions of no prior knowledge.
Detect Any Keypoints: An Efficient Light-Weight Few-Shot Keypoint Detector
Changsheng Lu (Australian National University), Piotr Koniusz (Data61 CSIRO)
Object DetectionPose EstimationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A lightweight few-shot keypoint detection model is proposed, integrating feature modulation and detection while combining mean feature contrastive learning to enhance cross-species generalization ability.
Detecting and Preventing Hallucinations in Large Vision Language Models
Anisha Gunjal (University of Texas), Erhan Bas (Scale AI)
OptimizationTransformerReinforcement LearningVision Language ModelImageMultimodality
🎯 What it does: This paper constructs the M-HalDetect multimodal hallucination detection dataset and trains reward models and direct optimization models based on this dataset to reduce the hallucination rate of large visual language models.
Detection and Defense of Unlearnable Examples
Yifan Zhu (Chinese Academy of Sciences), Xiao-Shan Gao (Chinese Academy of Sciences)
Anomaly DetectionAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This study investigates the detectability of unlearnable examples and defense methods, proving their linear separability and proposing two detection algorithms; subsequently, a defense scheme based on strong data augmentation and adversarial noise generated by simple networks is designed; and a theoretical boundary between toxicity budget and adversarial training budget is provided.
Detection-Based Intermediate Supervision for Visual Question Answering
Yuhang Liu (ByteDance Inc.), Dangyang Chen (Ping An Property & Casualty Insurance Company of China)
RecognitionObject DetectionExplainability and InterpretabilityTransformerVision Language ModelMultimodality
🎯 What it does: A detection-based intermediate supervision (DIS) method is proposed, which transforms intermediate inference results into serializable detection boxes, ground truths, and answer sequences. It utilizes an autoregressive decoder to supervise the inference state of the visual question answering model, improving answer accuracy and reasoning consistency.
Devignet: High-Resolution Vignetting Removal via a Dual Aggregated Fusion Transformer with Adaptive Channel Expansion
Shenghong Luo (University of Macau), Chi-Man Pun (University of Macau)
RestorationTransformerImage
🎯 What it does: A high-resolution vignetting dataset called VigSet has been constructed, and DeVigNet has been proposed for vignetting removal.
DexFuncGrasp: A Robotic Dexterous Functional Grasp Dataset Constructed from a Cost-Effective Real-Simulation Annotation System
Jinglue Hang (Dalian University of Technology), Yi Sun (Dalian University of Technology)
Data SynthesisPose EstimationRobotic IntelligenceReinforcement LearningPoint Cloud
🎯 What it does: A low-cost, high-efficiency real-time 'real-simulation' labeling system is proposed, which can directly generate and record the functional grasp postures of a high-degree-of-freedom robotic hand from human hand movements, and based on this, the first DexFuncGrasp dataset containing functional grasp postures and contact area annotations has been constructed.
DGA-GNN: Dynamic Grouping Aggregation GNN for Fraud Detection
Mingjiang Duan (Zhejiang University), Xinyu Wang (Zhejiang University)
Anomaly DetectionGraph Neural NetworkGraphTabularFinance Related
🎯 What it does: A dynamic grouping aggregation graph neural network (DGA-GNN) is proposed, which processes non-additive attributes through decision tree binning encoding and uses feedback-based dynamic grouping to bipartition neighboring nodes, ultimately achieving more discriminative fraud detection.
DGCLUSTER: A Neural Framework for Attributed Graph Clustering via Modularity Maximization
Aritra Bhowmick (New York University), Sourav Medya (University of Illinois)
Graph Neural NetworkGraph
🎯 What it does: Proposes the DGCLUSTER framework, which utilizes GNN to learn node embeddings and achieves modular maximization of graph clustering without the need to preset the number of clusters based on embedding similarity.
DGL: Dynamic Global-Local Prompt Tuning for Text-Video Retrieval
Xiangpeng Yang (University of Technology Sydney), Yi Yang (Zhejiang University)
RetrievalTransformerPrompt EngineeringVideoTextMultimodality
🎯 What it does: For the text-video retrieval task, a Dynamic Global-Local Prompt Tuning (DGL) framework is proposed, which utilizes a shared latent space to generate cross-modal prompts and incorporates global-local attention in the visual encoder to capture both overall and frame-level information of the video.
DGPO: Discovering Multiple Strategies with Diversity-Guided Policy Optimization
Wentse Chen (Carnegie Mellon University), Jun Zhu (Tsinghua University)
OptimizationReinforcement LearningSequential
🎯 What it does: A diversity-guided policy optimization algorithm DGPO is proposed, which learns multiple high-reward policies on a single network.
DHGCN: Dynamic Hop Graph Convolution Network for Self-Supervised Point Cloud Learning
Jincen Jiang (Northwest Agricultural and Forestry University), Meili Wang (Peking University)
ClassificationSegmentationRepresentation LearningGraph Neural NetworkContrastive LearningPoint Cloud
🎯 What it does: This paper studies a self-supervised Dynamic Hop Graph Convolutional Network (DHGCN) for point cloud learning.
DI-V2X: Learning Domain-Invariant Representation for Vehicle-Infrastructure Collaborative 3D Object Detection
Xiang Li (Beijing Institute of Technology), Jianbing Shen (University of Macau)
Object DetectionDomain AdaptationAutonomous DrivingKnowledge DistillationRepresentation LearningPoint Cloud
🎯 What it does: This paper proposes the DI-V2X model, which utilizes a teacher-student distillation framework to learn domain-invariant 3D object detection representations in vehicle-infrastructure collaborative perception.
Diagnosing and Rectifying Fake OOD Invariance: A Restructured Causal Approach
Ziliang Chen (Jinan University), Liang Lin (Sun Yat-sen University)
Domain AdaptationContrastive LearningImageBenchmark
🎯 What it does: This paper proposes the ReStructured SCM model by reconstructing the causal graph, and based on this, designs the IIL framework, which uses a soft feature selector to correct pseudo-invariant features in IRL.
Dialogue for Prompting: A Policy-Gradient-Based Discrete Prompt Generation for Few-Shot Learning
Chengzhengxu Li (Xi'an Jiaotong University), Chao Shen (Xi'an Jiaotong University)
ClassificationOptimizationTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: A new discrete prompt optimization framework DP2O is proposed, which utilizes multi-turn dialogue with GPT-4 to generate high-quality readable prompts, filtered through the SUE metric, and then employs a policy gradient-based RL to match the most suitable prompts for downstream tasks.
Dialogues Are Not Just Text: Modeling Cognition for Dialogue Coherence Evaluation
Xue Li (Harbin Institute of Technology), Yi Guan (University of Science and Technology of China)
Graph Neural NetworkTransformerText
🎯 What it does: A dialogue coherence evaluation framework DCGEval is proposed, based on Dialogue Cognitive Graph (DCG) and an information interaction enhancement module, simulating human dual-system decision-making and integrating four cognitive abilities: core semantics, time, roles, and common sense.
DiDA: Disambiguated Domain Alignment for Cross-Domain Retrieval with Partial Labels
Haoran Liu (Sichuan University), Xu Wang (Sichuan University)
RetrievalDomain AdaptationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes the 'Partial Label' problem in cross-domain image retrieval (PCIR) and designs a new method called DiDA, which utilizes Prototype Score Unit Learning (PSUL) and Prototype Domain Alignment (PBDA) to achieve label disambiguation and cross-domain feature alignment, thereby improving retrieval performance.
DifAttack: Query-Efficient Black-Box Adversarial Attack via Disentangled Feature Space
Jun Liu (University of Macau), Jinyu Tian (Beijing Normal University)
Adversarial AttackAuto EncoderImage
🎯 What it does: This paper proposes a black-box adversarial attack method called DifAttack based on feature space decoupling. It first uses an autoencoder to split the latent features of images into adversarial features and visual features. During the attack phase, only the adversarial features are optimized to generate adversarial samples while keeping the visual features unchanged, and finally controls the perturbation magnitude through projection.
DiffAIL: Diffusion Adversarial Imitation Learning
Bingzheng Wang (Shandong University), Yilong Yin (Shandong University)
Robotic IntelligenceReinforcement Learning from Human FeedbackReinforcement LearningDiffusion modelSequential
🎯 What it does: In an environment without an explicit reward function, the introduction of diffusion models into the adversarial imitation learning framework improves the distribution matching capability of the discriminator, thereby achieving higher quality imitation learning.
DiffBEV: Conditional Diffusion Model for Bird’s Eye View Perception
Jiayu Zou (Institute of Automation, Chinese Academy of Sciences), Xingang Wang (Institute of Automation, Chinese Academy of Sciences)
Object DetectionSegmentationAutonomous DrivingTransformerDiffusion modelImagePoint Cloud
🎯 What it does: An end-to-end DiffBEV framework is proposed, utilizing a conditional diffusion model for stepwise denoising and refinement of Bird's Eye View (BEV) features, thereby enhancing perception quality in autonomous driving.
Differentiable Auxiliary Learning for Sketch Re-Identification
Xingyu Liu (Nanjing University of Information Science and Technology), Guoying Zhao (University of Oulu)
RecognitionRetrievalConvolutional Neural NetworkContrastive LearningImageMultimodality
🎯 What it does: Designed and implemented a Differentiable Auxiliary Learning Network (DALNet), which generates background noise-free auxiliary modalities through a dynamic auxiliary generator, and enhances the cross-modal matching performance of Sketch Re-ID by combining a multimodal interaction attention module and collaborative learning.
DiffRAW: Leveraging Diffusion Model to Generate DSLR-Comparable Perceptual Quality sRGB from Smartphone RAW Images
Mingxin Yi (Tsinghua University), Jingduo Tian (Huawei)
Image TranslationGenerationDiffusion modelImage
🎯 What it does: A mobile RAW to DSLR-level sRGB image conversion framework called DiffRAW is developed, utilizing RAW structural information and color position preservation conditions for conditional generation, and designing a Domain Transform Diffusion Method to enhance generated details and significantly reduce inference steps.
DiffSED: Sound Event Detection with Denoising Diffusion
Swapnil Bhosale (University of Surrey), Xiatian Zhu (Imperial College London)
ClassificationRecognitionGenerationTransformerDiffusion modelAudio
🎯 What it does: This paper proposes DiffSED, a sound event detection framework based on a denoising diffusion model, which utilizes noise latent queries to progressively denoise in the Transformer decoder, generating event time boundaries and category labels.
Diffusion Language-Shapelets for Semi-supervised Time-Series Classification
Zhen Liu (South China University of Technology), Qianli Ma (South China University of Technology)
ClassificationExplainability and InterpretabilityDiffusion modelContrastive LearningTime Series
🎯 What it does: The DiffShape model is proposed and implemented, which combines self-supervised diffusion learning with language contrastive learning to generate interpretable shape subsequences (shapelets) and is trained on semi-supervised time series classification tasks.
DiffusionEdge: Diffusion Probabilistic Model for Crisp Edge Detection
Yunfan Ye (Hunan University), Zhiping Cai (National University of Defense Technology)
SegmentationKnowledge DistillationDiffusion modelAuto EncoderImage
🎯 What it does: A diffusion probability model-based edge detector called DiffusionEdge is proposed, which can generate accurate and sharp edge maps directly at the original resolution without the need for post-processing.
DiffusionTrack: Diffusion Model for Multi-Object Tracking
Run Luo (Shenzen Institute of Advanced Technology), Min Yang (Huazhong University of Science and Technology)
Object DetectionObject TrackingTransformerDiffusion modelVideo
🎯 What it does: This paper proposes DiffusionTrack, a multi-object tracking framework that unifies object detection and association into a denoising diffusion process for tracking.
DiG-In-GNN: Discriminative Feature Guided GNN-Based Fraud Detector against Inconsistencies in Multi-Relation Fraud Graph
Jinghui Zhang (Southeast University), Fang Dong (Southeast University)
Anomaly DetectionGraph Neural NetworkReinforcement LearningContrastive LearningGraphFinance Related
🎯 What it does: This paper proposes DiG-In-GNN, which generates distinguishable guiding nodes through multi-scale contrastive learning and uses reinforcement learning for fine-grained neighbor selection, enhancing the effectiveness of multi-relational graph fraud detection.
DINGO: Towards Diverse and Fine-Grained Instruction-Following Evaluation
Zihui Gu (Renmin University of China), Ju Fan (Renmin University of China)
Large Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: A multi-level, fine-grained evaluation dataset DINGO has been constructed, covering diverse user instructions, and various LLMs have been evaluated using LLM-as-a-judge.
Direct Amortized Likelihood Ratio Estimation
Adam D. Cobb (SRI International), Susmit Jha (SRI International)
OptimizationReinforcement LearningTabularBenchmark
🎯 What it does: A neural network estimator for directly calculating the likelihood ratio between parameter pairs (DNRE) is proposed, and based on this, a Monte-Carlo posterior approximation and numerically stable gradient estimation are derived to compare Hamiltonian Monte Carlo (HMC) and random walk Metropolis-Hastings (MH) in simulating likelihood-free inference, with the method applied to the quadrotor design problem.
Direct May Not Be the Best: An Incremental Evolution View of Pose Generation
Yuelong Li (Tiangong University), Jianming Wang (Tiangong University)
GenerationPose EstimationRecurrent Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: A fine-grained incremental evolution framework is proposed for human pose generation, capable of outputting a series of intermediate poses.
Directed Diffusion: Direct Control of Object Placement through Attention Guidance
Wan-Duo Kurt Ma (Victoria University of Wellington), W. Bastiaan Kleijn (Google Research)
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: A training-free method based on cross-attention mapping is proposed—Directed Diffusion, which allows for precise control of the positions of multiple objects in text-to-image diffusion models.
Direction-Aware Video Demoiréing with Temporal-Guided Bilateral Learning
Shuning Xu (University of Macau), Jiantao Zhou (University of Macau)
RestorationConvolutional Neural NetworkVideo
🎯 What it does: A dual-stage DTNet network is proposed, utilizing direction-aware DCT and time-guided bilateral learning to remove moiré patterns, align features, correct colors, and restore details in videos.
Dirichlet-Based Prediction Calibration for Learning with Noisy Labels
Chen-Chen Zong (Nanjing University of Aeronautics and Astronautics), Sheng-Jun Huang (Nanjing University of Aeronautics and Astronautics)
ClassificationData-Centric LearningContrastive LearningImage
🎯 What it does: A Dirichlet-based prediction calibration method (DPC) is proposed, which reduces the overconfidence problem in learning from noisy labels and improves example selection effectiveness by modifying softmax and training with the Dirichlet distribution.
Discerning Temporal Difference Learning
Jianfei Ma (Northwestern Polytechnical University)
Reinforcement LearningTabular
🎯 What it does: This paper proposes a tunable temporal difference learning algorithm called DTD, which reallocates the TD error of historical states using an emphasis function.
DiSCO: Diffusion Schrödinger Bridge for Molecular Conformer Optimization
Danyeong Lee (Seoul National University), Sun Kim (Seoul National University)
OptimizationDrug DiscoveryDiffusion modelTabularStochastic Differential Equation
🎯 What it does: Proposes the DiSCO framework, which utilizes the diffusion Schrödinger bridge for post-processing optimization of existing molecular conformations, making the generated 3D conformations closer to the true energy distribution.
Discovering Sequential Patterns with Predictable Inter-event Delays
Joscha Cüppers, Jilles Vreeken (CISPA Helmholtz Center for Information Security)
Time SeriesSequential
🎯 What it does: A HOPPER algorithm based on the Minimum Description Length (MDL) principle is proposed, specifically for mining serial patterns with predictable inter-event delays in sequential data.
Discrepancy and Uncertainty Aware Denoising Knowledge Distillation for Zero-Shot Cross-Lingual Named Entity Recognition
Ling Ge (Beihang University), Hong Zhang (National Computer Network Emergency Response Technical Team)
RecognitionKnowledge DistillationText
🎯 What it does: The DenKD model is proposed, which uses denoising knowledge distillation to address the issue of pseudo-label noise in zero-shot cross-lingual named entity recognition.
Discrete Cycle-Consistency Based Unsupervised Deep Graph Matching
Siddharth Tourani (Heidelberg University), Bogdan Savchynskyy (MBZUAI)
RecognitionOptimizationGraph Neural NetworkImage
🎯 What it does: This paper proposes an unsupervised deep graph matching framework based on discrete cycle consistency and black-box differentiation for image keypoint matching.
Discretization-Induced Dirichlet Posterior for Robust Uncertainty Quantification on Regression
Xuanlong Yu (Paris-Saclay University), Emanuel Aldea (University of Oxford)
Depth EstimationSuper ResolutionImage
🎯 What it does: This study proposes a general auxiliary uncertainty estimator (AuxUE) that separates model uncertainty into observable uncertainty and known uncertainty, achieving robust quantification in regression tasks.
Discriminative Forests Improve Generative Diversity for Generative Adversarial Networks
Junjie Chen (Harbin Institute of Technology), Xinghua Shi (Temple University)
GenerationData SynthesisGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes the Discriminative Forest GAN (ForestGAN), which constructs a forest of multiple independent discriminators using a bootstrap method to enhance the diversity of GAN generation.
Discriminatively Fuzzy Multi-View K-means Clustering with Local Structure Preserving
Jun Yin (Shanghai Maritime University), Pei Wang (Shanghai Maritime University)
OptimizationMultimodality
🎯 What it does: A multi-view fuzzy K-means clustering method, called DFMKLS, is proposed, which takes into account the internal compactness of clusters, the separability between different clusters, and the preservation of local structure.
Disentangled Diffusion-Based 3D Human Pose Estimation with Hierarchical Spatial and Temporal Denoiser
Qingyuan Cai (Beijing Normal University), Yongzhen Huang (Beijing Normal University)
Pose EstimationTransformerDiffusion modelImage
🎯 What it does: A 3D human pose estimation method based on diffusion models, DDHPose, is proposed, which utilizes bone lengths and bone vectors for decoupled diffusion, and enhances the hierarchical relationship of bones through a Hierarchical Space-Time Denoiser (HSTDenoiser) in the reverse process.
Disentangled Partial Label Learning
Wei-Xuan Bao (Southeast University), Min-Ling Zhang (Lenovo Group Ltd.)
ClassificationRepresentation LearningGraph Neural NetworkImage
🎯 What it does: This paper proposes a partial label learning method called TERIAL based on decomposable representation learning, which achieves factor-level separation of instances and labels through an instance-label bipartite graph and a neighborhood routing mechanism for classification prediction.
Disguise without Disruption: Utility-Preserving Face De-identification
Zikui Cai (University of California), Ziyan Wu (United Imaging Intelligence)
RecognitionData SynthesisSafty and PrivacyMixture of ExpertsAuto EncoderGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a facial de-identification method called Disguise, which can anonymize faces without compromising the practical attributes of the images.
Disjoint Partial Enumeration without Blocking Clauses
Giuseppe Spallitta (University of Trento), Armin Biere (University of Freiburg)
🎯 What it does: A non-blocking clause disjoint AllSAT solver called TABULARALLSAT is designed to enumerate mutually exclusive partial models.
Distilling Autoregressive Models to Obtain High-Performance Non-autoregressive Solvers for Vehicle Routing Problems with Faster Inference Speed
Yubin Xiao (Jilin University), You Zhou (Jilin University)
OptimizationComputational EfficiencyKnowledge DistillationTransformerTabular
🎯 What it does: Transforming a Transformer-based autoregressive (AR) model into a non-autoregressive (NAR) model through knowledge distillation for fast inference in vehicle routing problems (VRP).
Distilling Reliable Knowledge for Instance-Dependent Partial Label Learning
Dong-Dong Wu (Southeast University), Min-Ling Zhang (Southeast University)
ClassificationKnowledge DistillationRepresentation LearningContrastive LearningImage
🎯 What it does: A self-distillation-based framework DIRK is proposed for instance-dependent partial label learning (IDPLL), and a representation refinement module DIRK-REF is added to enhance feature representation and classification performance.
DistilVPR: Cross-Modal Knowledge Distillation for Visual Place Recognition
Sijie Wang (Nanyang Technological University), Wee Peng Tay (Nanyang Technological University)
RecognitionRetrievalKnowledge DistillationImageMultimodalityPoint Cloud
🎯 What it does: This paper proposes DistilVPR—a cross-modal knowledge distillation pipeline that enhances the performance of unimodal students in visual place recognition by utilizing multi-agent and multi-manifold relationships.
Distributed Manifold Hashing for Image Set Classification and Retrieval
Xiaobo Shen (Nanjing University of Science and Technology), Yuhui Zheng (Nanjing University of Information Science and Technology)
ClassificationRetrievalImage
🎯 What it does: A Distributed Manifold Hashing (DMH) method is proposed for the classification and retrieval of distributed image sets, capable of training and inference on multi-node networks.
Distribution Matching for Multi-Task Learning of Classification Tasks: A Large-Scale Study on Faces & Beyond
Dimitrios Kollias (Queen Mary University of London), Stefanos Zafeiriou (Imperial College London)
ClassificationKnowledge DistillationImage
🎯 What it does: In multi-task learning, the authors propose a knowledge exchange mechanism that transforms the prior correlations between tasks into two coupling losses through distribution matching and soft co-labeling, enabling model training in scenarios where task annotations do not overlap or where there are significant differences in sample sizes.
Distribution-Conditioned Adversarial Variational Autoencoder for Valid Instrumental Variable Generation
Xinshu Li (University of New South Wales), Lina Yao (CSIRO Data61)
GenerationData SynthesisAuto EncoderGenerative Adversarial NetworkTabular
🎯 What it does: The study proposes a Variational Inference with Variational Adversarial Networks (VIV) to automatically generate instrumental variables that satisfy the conditions of relevance, exclusivity, and exogeneity in the absence of effective candidate instrumental variables, to support causal inference.
Distributional Off-Policy Evaluation for Slate Recommendations
Shreyas Chaudhari (University of Massachusetts), Nikos Vlassis (Adobe)
Recommendation SystemReinforcement LearningTabular
🎯 What it does: A method for unbiased distribution estimation for slide recommendation (SUnO) is proposed, which can estimate the reward distribution of the target policy based on offline data.
DIUSum: Dynamic Image Utilization for Multimodal Summarization
Min Xiao (Institute of Automation, Chinese Academy of Sciences), Chengqing Zong (Institute of Automation, Chinese Academy of Sciences)
TransformerImageTextMultimodality
🎯 What it does: A dynamic image utilization framework (DIUSum) is proposed, which evaluates the effectiveness of images through an image selector and dynamically injects valid image information during the decoding phase to enhance the quality of multimodal summaries.
Divergence-Guided Simultaneous Speech Translation
Xinjie Chen (Zhejiang University), Zhongqiang Huang (Alibaba DAMO Academy)
TransformerAudio
🎯 What it does: This paper proposes the DiG-SST framework, which combines a prefix-enhanced end-to-end speech translation model with a dynamic read-write strategy based on normalized cosine divergence, achieving low-latency and high-quality synchronous speech translation.
Diverse and Aligned Audio-to-Video Generation via Text-to-Video Model Adaptation
Guy Yariv (Hebrew University of Jerusalem), Yossi Adi (Hebrew University of Jerusalem)
GenerationData SynthesisDiffusion modelVideoTextMultimodalityAudio
🎯 What it does: Based on a pre-trained text-to-video diffusion model, a lightweight adapter network is used to map audio encoding to pseudo-text tokens, enabling the generation of diverse and realistic videos driven by natural audio, and for the first time supporting audio + text dual-modal generation.
Diverse and Stable 2D Diffusion Guided Text to 3D Generation with Noise Recalibration
Xiaofeng Yang (Nanyang Technological University), Guosheng Lin (Nanyang Technological University)
GenerationData SynthesisDiffusion modelScore-based ModelNeural Radiance FieldImageText
🎯 What it does: The Noise Recalibration SDS (NR-SDS) algorithm is proposed, which improves the training of NeRF based on a 2D diffusion model through single noise training and noise recalibration loss, achieving high-quality and diverse 3D generation.
Diverse Person: Customize Your Own Dataset for Text-Based Person Search
Zifan Song (Tongji University), Cairong Zhao (Tongji University)
RetrievalLarge Language ModelDiffusion modelImageText
🎯 What it does: The Diverse Person (DP) framework is proposed, which enhances the text retrieval dataset for person search by editing the original data using a diffusion model and generating high-quality text with LLM.
Diversity-Authenticity Co-constrained Stylization for Federated Domain Generalization in Person Re-identification
Fengxiang Yang (Xiamen University), Nicu Sebe (Reconova Technologies)
RecognitionDomain AdaptationFederated LearningConvolutional Neural NetworkTransformerImage
🎯 What it does: In the context of federated learning, the DACS method is proposed for the domain generalization task of person re-identification (re-ID), utilizing a style transfer model to generate diverse and realistic synthetic samples, thereby enhancing the generalization ability of local models.
Divide and Conquer: Hybrid Pre-training for Person Search
Yanling Tian (Nanjing University of Science and Technology), Shanshan Zhang (Nanjing University of Science and Technology)
RecognitionObject DetectionRetrievalDomain AdaptationContrastive LearningImage
🎯 What it does: A hybrid pre-training framework for person search is proposed, which utilizes data from detection and re-identification sub-tasks for joint learning and reduces domain differences through a domain alignment module.
DLCA-Recon: Dynamic Loose Clothing Avatar Reconstruction from Monocular Videos
Chunjie Luo (Wuhan University), Chunxia Xiao (Wuhan University)
GenerationPose EstimationConvolutional Neural NetworkVideo
🎯 What it does: This paper proposes a method called DLCA-Recon, which can reconstruct 3D human figures with dynamic loose clothing from monocular videos.
DME: Unveiling the Bias for Better Generalized Monocular Depth Estimation
Songsong Yu (Dalian University of Technology), Huchuan Lu (Dalian University of Technology)
Depth EstimationConvolutional Neural NetworkTransformerMixture of ExpertsImage
🎯 What it does: Analyzes the long-tail distribution and its correlation with simulation in monocular depth estimation, and proposes a Distance-based Multi-Expert (DME) network that achieves depth prediction fusion across different distance segments through pixel-level routing; simultaneously designs a two-stage training strategy and experimentally validates its performance across multiple datasets.
DMMR: Cross-Subject Domain Generalization for EEG-Based Emotion Recognition via Denoising Mixed Mutual Reconstruction
Yiming Wang (Xi'an Jiaotong University), Yujiao Tang (Xi'an Jiaotong University)
RecognitionDomain AdaptationAuto EncoderTime SeriesBiomedical Data
🎯 What it does: A self-supervised mixed mutual reconstruction (DMMR) model is proposed for cross-subject domain generalization in EEG emotion recognition.
DocFormerv2: Local Features for Document Understanding
Srikar Appalaraju (Amazon Web Services), R. Manmatha (Amazon Web Services)
TransformerTextMultimodality
🎯 What it does: This paper proposes DocFormerv2, a multimodal encoder-decoder Transformer for visual document understanding (VDU) that enhances the perception of document layout through pre-training on local features.
DocMSU: A Comprehensive Benchmark for Document-Level Multimodal Sarcasm Understanding
Hang Du (Beijing University of Posts and Telecommunications), Xudong Jiang (Nanyang Technological University)
ClassificationObject DetectionTransformerVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: This paper proposes the DocMSU document-level multimodal sarcasm understanding benchmark and designs a pixel-level image and word-level text fine-grained alignment sliding window Swin-Transformer fusion model to achieve sarcasm detection and localization in news text and accompanying images.
DocNLC: A Document Image Enhancement Framework with Normalized and Latent Contrastive Representation for Multiple Degradations
Ruilu Wang (South China University of Technology), Lianwen Jin (South China University of Technology)
RestorationRepresentation LearningConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A unified document image enhancement framework, DocNLC, is proposed, which achieves unified processing of various degradation types through normalization and latent contrastive learning.
DOCTR: Disentangled Object-Centric Transformer for Point Scene Understanding
Xiaoxuan Yu (Samsung Research China), Younghun Sung (Samsung Advanced Institute of Technology)
Object DetectionSegmentationPose EstimationTransformerPoint Cloud
🎯 What it does: This paper proposes a Disentangled Object-Centric Transformer (DOCTR) that simultaneously performs instance segmentation, pose estimation, and mesh reconstruction of point cloud scenes through a single network, addressing the challenges of modeling and optimizing inter-object relationships in traditional multi-stage pipelines.
Does Few-Shot Learning Suffer from Backdoor Attacks?
Xinwei Liu (Institute of Information Engineering, Chinese Academy of Sciences), Xiaochun Cao (Sun Yat-sen University)
Adversarial AttackMeta LearningConvolutional Neural NetworkImage
🎯 What it does: This paper studies backdoor attacks in few-shot learning (FSL) and proposes an FLBA attack method based on maximizing embedding bias triggers and hidden perturbations.
DOGE-Train: Discrete Optimization on GPU with End-to-End Training
Ahmed Abbas (Max Planck Institute for Informatics), Paul Swoboda (Max Planck Institute for Informatics)
OptimizationGraph Neural NetworkTransformerGraph
🎯 What it does: A differentiable algorithm that combines graph neural networks and Lagrangian decomposition is proposed for efficiently solving the LP relaxation of 0-1 integer linear programming.
Domain Generalizable Person Search Using Unreal Dataset
Minyoung Oh (UNIST), Jae-Young Sim (UNIST)
Object DetectionDomain AdaptationContrastive LearningImage
🎯 What it does: A domain generalization framework for person search is designed, which is trained using only automatically annotated virtual data and can be directly tested on any unseen real data.
Domain Generalization with Vital Phase Augmentation
Ingyun Lee (Korea Advanced Institute of Science and Technology), Hyun Myung (Korea Advanced Institute of Science and Technology)
Domain AdaptationConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: This paper proposes Vital Phase Augmentation (VIPAug), which enhances the model's robustness to input distortion and phase fluctuations by performing limited transformations on the image phase in the frequency domain and combining it with amplitude enhancement.
Domain Invariant Learning for Gaussian Processes and Bayesian Exploration
Xilong Zhao (Shanghai Jiao Tong University), Nanyang Ye (Shanghai Jiao Tong University)
Domain AdaptationOptimizationTabular
🎯 What it does: This paper proposes a domain-invariant learning algorithm for Gaussian processes, DIL-GP, which enhances its generalization ability to out-of-domain data by adaptively partitioning data and performing a min-max optimization of the IRM penalty, and extends it to Bayesian optimization.
Domain-Aware Fine-Tuning: Enhancing Neural Network Adaptability
Seokhyeon Ha (Seoul National University), Jungwoo Lee (Seoul National University)
ClassificationSegmentationDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningContrastive LearningImage
🎯 What it does: A new fine-tuning method is proposed—Domain-Aware Fine-Tuning (DAFT), which reduces the distortion of pre-trained features and enhances adaptability through batch normalization (BN) transformation combined with one-time linear probing (LP) and fine-tuning (FT).
Domain-Controlled Prompt Learning
Qinglong Cao (Shanghai Jiao Tong University), Xiaokang Yang (Shanghai Jiao Tong University)
RecognitionDomain AdaptationTransformerPrompt EngineeringContrastive LearningImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: By introducing a specific domain foundation model (LSDM) and conducting domain-controlled prompt learning on the visual and language branches, the zero-shot recognition performance of CLIP on remote sensing and medical images is improved.
Domain-Hallucinated Updating for Multi-Domain Face Anti-spoofing
Chengyang Hu (Shanghai Jiao Tong University), Lizhuang Ma (East China Normal University)
RecognitionDomain AdaptationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: Proposes the Domain-Hallucinated Updating (DHU) framework, which updates the face anti-spoofing model by generating pseudo 'old domain' features without the previous domain data, balancing new domain performance and old domain knowledge retention.
Double Auction on Diffusion Network
Miao Li (ShanghaiTech University), Dengji Zhao (ShanghaiTech University)
Graph
🎯 What it does: A double auction mechanism DTR suitable for social networks is proposed, utilizing dynamic trade reduction and invitation incentives to ensure honest reporting from both parties and avoid market deficits, while encouraging traders to invite neighbors to join.
Double Buffers CEM-TD3: More Efficient Evolution and Richer Exploration
Sheng Zhu (Jilin University), Daolong An (Jilin University)
Reinforcement Learning
🎯 What it does: This paper proposes Double Buffers CEM-TD3, which improves the evolutionary and gradient learning process of the original CEM-TD3 using a double buffering mechanism, enhancing evolutionary efficiency and population diversity.
Double-Bounded Optimal Transport for Advanced Clustering and Classification
Liangliang Shi (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
ClassificationOptimizationImage
🎯 What it does: Proposes Dual-Bound Optimal Transport (DB-OT) and applies it to improve the loss and inference of centroid clustering and controllable cluster size in long-tail classification.
Double-Descent Curves in Neural Networks: A New Perspective Using Gaussian Processes
Ouns El Harzli (University of Oxford), Ard A. Louis (University of Oxford)
🎯 What it does: Using random matrix theory and NNGP, the impact of width on the double descent curve is analyzed, and analytical expressions for width-related random kernels and spectral distributions are provided.
Double-Layer Hybrid-Label Identification Feature Selection for Multi-View Multi-Label Learning
Pingting Hao (Jilin University), Wanfu Gao (Portland State University)
ClassificationOptimizationMultimodality
🎯 What it does: A dual-layer hybrid label identification (DHLI) framework is proposed for multi-view multi-label feature selection, which first splits the observed labels into common labels, view-specific labels, and noise labels, and then jointly optimizes feature weights, label decomposition, and noise suppression to ultimately obtain a high-quality feature sequence.
Doubly Perturbed Task Free Continual Learning
Byung Hyun Lee (Seoul National University), Se Young Chun (Seoul National University)
ClassificationOptimizationImage
🎯 What it does: A dual perturbation task-agnostic continual learning (DPCL) framework is proposed, which achieves prospective learning of future samples by simultaneously applying perturbations to both inputs and weights.
DP-AdamBC: Your DP-Adam Is Actually DP-SGD (Unless You Apply Bias Correction)
Qiaoyue Tang (University of British Columbia), Mathias Lécuyer (University of British Columbia)
OptimizationSafty and PrivacyGraph Neural NetworkImageTextGraph
🎯 What it does: An optimization algorithm called DP-AdamBC is proposed to correct the bias of DP Adam, restoring the original behavior of Adam under privacy training;
DPA-P2PNet: Deformable Proposal-Aware P2PNet for Accurate Point-Based Cell Detection
Zhongyi Shui (Zhejiang University), Lin Yang (Westlake University)
Object DetectionSegmentationConvolutional Neural NetworkContrastive LearningImageBiomedical Data
🎯 What it does: An end-to-end point-based cell detection model DPA-P2PNet is proposed, supporting multi-scale decoding, deformable point candidates, and multi-field input, and for the first time, self-supervised pre-training is conducted on a large-scale immunohistochemical image dataset.
DR-Label: Label Deconstruction and Reconstruction of GNN Models for Catalysis Systems
Bowen Wang (Chinese University of Hong Kong), Pheng Ann Heng
Graph Neural NetworkGraph
🎯 What it does: This paper studies a graph neural network supervision and prediction strategy named DR-Label, aimed at reducing the multiplicity of edge representations and sensitivity to changes in graph structure in catalyst-adsorbate systems.
DreamIdentity: Enhanced Editability for Efficient Face-Identity Preserved Image Generation
Zhuowei Chen (University of Science and Technology of China), Zhendong Mao (ByteDance)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: Designed and implemented a non-optimized face identity-preserving generation method called DreamIdentity, which utilizes a multi-word multi-scale ID encoder to map a single face into the word embedding space, and enhances editability through self-enhanced editable learning, achieving high-quality, editable, and identity-preserving image generation.
DreamStyler: Paint by Style Inversion with Text-to-Image Diffusion Models
Namhyuk Ahn (NAVER WEBTOON AI), Kibeom Hong (SwatchOn)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: We propose DreamStyler, a framework that enables text generation and style transfer using a single style reference image, supporting instant artistic creation;
DRF: Improving Certified Robustness via Distributional Robustness Framework
Zekai Wang (Wuhan University), Weiwei Liu (Wuhan University)
ClassificationOptimizationAdversarial AttackConvolutional Neural NetworkGaussian SplattingImage
🎯 What it does: This paper proposes a Distributed Robustness Framework (DRF) that unifies adversarial training-based random smoothing methods with distributional robustness theory. It introduces a learnable soft projection term during the training process, allowing for adaptive adjustment of the distance between adversarial samples and original samples, thereby enhancing the average certified radius (ACR) of the smooth classifier and the certified accuracy at different radii.
DrFuse: Learning Disentangled Representation for Clinical Multi-Modal Fusion with Missing Modality and Modal Inconsistency
Wenfang Yao (Hong Kong Polytechnic University), Jing Qin (Hong Kong Polytechnic University)
ClassificationRepresentation LearningTransformerMultimodalityTime SeriesBiomedical DataElectronic Health Records
🎯 What it does: The DrFuse model is designed to utilize decoupled shared/distinct feature learning to fuse electronic health records and chest X-ray images, addressing the issues of missing modalities and modality inconsistency through an attention mechanism, thereby improving clinical prediction accuracy.